TL;DR: Every attribution model answers the same question: which touchpoints were present when a conversion happened? None of them answer the question that actually matters for budget decisions: which touchpoints caused the conversion? That gap — between correlation and causation — is where most B2B marketing budgets quietly bleed out.
What Attribution Models Actually Measure
Attribution is a bookkeeping system. It observes which marketing touchpoints appeared before a conversion and assigns credit to some combination of them. The touchpoints get credit. The conversion gets explained. The dashboard looks tidy.
What attribution cannot tell you:
- Whether the conversion would have happened without that touchpoint
- Whether a user in the holdout group (who didn’t see the ad) would have converted anyway
- How two channels interact over a multi-month buying cycle
- What drove the first-ever brand impression that made all later touchpoints work
The core limitation isn’t a flaw in any particular model. It’s structural. Attribution observes sequences of events and infers causation from them. But correlation is not causation — and in B2B marketing, the gap between them is enormous.
The Four Models and What They Miss
Last-touch attribution gives 100% credit to the final touchpoint before conversion. It’s the simplest model and the most misleading. It systematically rewards the channel that closes demand rather than the channels that create it. Brand campaigns, content marketing, and top-of-funnel paid social — which do the heavy lifting of building awareness — get zero credit. Branded search and retargeting, which intercept demand that already exists, capture most of the budget.
First-touch attribution has the inverse problem. It overcredits the initial touchpoint and ignores everything that nurtured the buyer from first awareness to purchase-ready. Useful for understanding where buyers come from. Useless for optimising a full-funnel budget.
Linear (multi-touch) attribution spreads credit evenly across all touchpoints. This is more honest than single-touch models in that it acknowledges a buyer journey exists. But “equal credit” is an arbitrary assumption. There’s no reason to believe that a display impression 90 days before purchase and a retargeting click 10 minutes before purchase contributed equally to the outcome.
Data-driven attribution sounds scientific because it uses machine learning. In practice, most implementations are still measuring correlation — they’re just measuring it more precisely. DDA identifies which touchpoint patterns are statistically associated with conversions. It cannot determine whether removing any touchpoint would change the conversion rate.
The core problem: Attribution models reward channels that are present when buyers convert. They do not measure whether those channels caused buyers to convert. These are fundamentally different questions.
The Selection Problem
The most important concept for understanding why attribution fails is the selection problem.
Retargeting campaigns target people who visited your website. These are not a random sample of the market — they’re people who already showed enough interest to navigate to your site. They were more likely to convert before they ever saw your retargeting ad.
When they convert at a high rate after seeing the retargeting ad, attribution gives the ad full credit. But the honest question is: would they have converted anyway?
In incrementality tests (randomised holdout experiments), the answer is usually: yes, most of them would have. Typically 60–80% of conversions attributed to retargeting would have occurred without the ad exposure. The incremental impact of the retargeting is far smaller than the attribution model suggests.
The selection problem applies everywhere:
- Branded search captures people who were already going to buy and just needed to find the website
- Email nurture sequences reach people who opted in because they were already interested
- Bottom-of-funnel content is read by people who are already far along in the decision process
All of these channels look highly effective in attribution because they intercept ready-to-convert buyers. None of them necessarily created that readiness.
The Efficiency Spiral
Attribution errors have a compounding effect on budget allocation.
When last-touch or multi-touch attribution shows that branded search, retargeting, and bottom-of-funnel content are “working” (high ROAS, low attributed CPA), budgets flow toward those channels. Meanwhile, brand campaigns, content marketing, and top-of-funnel awareness investment show poor attribution metrics — so they get cut.
The short-term effect looks like efficiency improvement. CPAs drop. ROAS rises. The dashboard looks better.
The long-term effect is that you’ve depleted the brand equity that made the bottom-of-funnel channels work. Over time, branded search volume declines. Retargeting audiences shrink. Conversion rates on bottom-of-funnel campaigns drop because fewer buyers are arriving with pre-existing brand awareness. You need to spend more on paid acquisition to achieve the same pipeline. CAC rises.
This is the Efficiency Spiral: optimizing for what’s easiest to measure starves the investments that make everything else efficient. Read more about the Efficiency Spiral →
What Honest Measurement Looks Like
The alternative to attribution isn’t no measurement. It’s measurement that’s designed to answer causal questions.
Incrementality testing (holdout experiments) directly measures whether a channel is causing conversions or just observing them. By randomly holding back a subset of the audience from a campaign and measuring conversion rates in both the exposed and holdout groups, you get a direct estimate of incremental impact. See What is Incrementality in Marketing? for a full explanation.
Marketing Mix Modelling (MMM) uses statistical regression across time-series data to estimate the contribution of each channel to overall sales — including offline channels, lagged effects, and interactions between channels. MMM is more complex and resource-intensive than holdout testing, but it captures dynamics that individual-campaign incrementality tests miss.
Brand tracking studies measure aided and unaided awareness, purchase intent, and consideration rates over time in target audiences. These connect brand investment to pipeline indirectly, through the mechanisms by which brand creates demand.
None of these tools are perfect. But they all share one quality that attribution lacks: they’re designed to answer causal questions, not just observational ones.
What to Do With Your Attribution Data
Attribution isn’t worthless. It’s excellent for some things:
- Understanding which channels are reaching buyers during the purchase journey
- Identifying friction points in the conversion path
- Monitoring relative changes in channel performance over time
- Informing where to place incremental spend at the margin
What it should never be used for: making large budget allocation decisions based on the assumption that attributed ROAS represents incremental value. It doesn’t. The ROAS figure your platform reports is almost always materially higher than the true incremental return.
The practical implication: run holdout tests on your highest-spend channels. Not once — quarterly. Build up an evidence base over time. Use the incrementality data to inform budget allocation. Use attribution data for what it’s actually good at: journey mapping and tactical optimisation.
FAQ
Q: If attribution is so flawed, why do platforms still report it?
Because it tells a flattering story. Every platform reports attribution metrics that maximise credit for that platform’s own channels. Google’s attribution model gives more credit to Google touchpoints. Meta’s gives more credit to Meta touchpoints. These are not neutral measurement tools — they are sales materials.
Q: What’s the simplest first step to move beyond attribution?
Run a holdout test on your highest-spend retargeting campaign. Most platforms (Google, Meta, LinkedIn) have built-in holdout functionality. A 15% holdout over 4–6 weeks will tell you whether the campaign is generating incremental conversions or just intercepting people who were going to convert anyway. One test will change how you see your data.
Q: Does this mean brand campaigns are always more valuable than performance campaigns?
No. It means the relative value of brand vs. performance investment is different from what attribution suggests — and the only way to know how different is to measure incrementally. Some performance campaigns are highly incremental. Some brand campaigns produce no measurable lift. The measurement is what tells you which is which.
Additional Resources
From the Zaitz Marketing Knowledge Library:
- What is Incrementality in Marketing? — The alternative to attribution that answers causal questions
- Why Your Customer Acquisition Cost Is Probably Wrong — How attribution errors flow through to broken CAC calculations
- Brand vs. Performance Is a False Choice — How the tagging wars create the conditions for the Efficiency Spiral
External Reading:
- How Meta’s Conversion API Works — Understanding server-side tracking as an attribution supplement
- Google’s Explanation of Attribution Models in GA4 — The official description of GA4’s data-driven model and its limitations ���
Want a second read on your measurement setup?
Start with a Growth Architecture Review. We will map your channel mix, audit your attribution, and show you where the real leverage is.
Book a Conversation →